TL;DR
This paper introduces ACID, a comprehensive, open-source MATLAB toolbox for processing and modeling diffusion MRI (dMRI) data from the brain, spinal cord, and ex vivo samples. ACID integrates advanced artifact correction, diffusion modeling (including DKI), and biophysical models, all within the widely-used SPM framework, aiming to standardize and streamline dMRI analysis.
What Problem Does This Paper Address?
Diffusion MRI (dMRI) is a powerful tool for understanding tissue microstructure, with growing applications in neuroscience and clinical settings. However, analyzing dMRI data is complex due to artifacts (motion, eddy currents, susceptibility) and the need for sophisticated modeling to extract meaningful biological information. While many tools exist, there's a need for a unified, comprehensive solution that handles diverse dMRI data (brain, spinal cord, ex vivo), incorporates state-of-the-art artifact correction and advanced modeling (like biophysical models), and integrates seamlessly with existing neuroimaging analysis pipelines. Specifically, spinal cord dMRI presents unique challenges due to lower signal-to-noise ratios (SNR) and different artifact patterns, and the validation of biophysical models often requires ex vivo data.
Methods
ACID is built as an extension of the Statistical Parametric Mapping (SPM) software, leveraging its robust infrastructure. Its core functionalities are organized into six modules:
- Startup: Handles data organization, conversion (DICOM to NIfTI), and metadata creation, adhering to the Brain Imaging Data Structure (BIDS) standard.
- Preprocessing: Implements advanced artifact correction techniques:
- ECMOCO: Eddy-current and motion correction, offering flexible registration options (volume-wise, slice-wise) tailored for different data types.
- msPOAS: Adaptive denoising to reduce noise while preserving tissue boundaries.
- RBC: Rician bias correction to address intensity biases in low-SNR data.
- HySCO: Susceptibility artifact correction to unwarp geometric distortions.
- Diffusion tensor/kurtosis imaging: Implements models like Diffusion Tensor Imaging (DTI) and Diffusion Kurtosis Imaging (DKI), offering various fitting algorithms (OLS, WLS, Robust, NLLS).
- Biophysical models: Includes models like WMTI-Watson and NODDI-DTI to estimate microstructural properties of white matter.
- Utilities: Provides tools for image manipulation, mask generation, quality assessment, and more.
- External tools: Allows integration of other established dMRI processing tools.
ACID emphasizes modularity, allowing users to build custom processing pipelines and benefit from SPM's segmentation, spatial normalization, and statistical analysis capabilities.
Datasets
The paper uses three example datasets to demonstrate ACID's capabilities:
- In vivo brain dMRI: From a healthy volunteer.
- In vivo spinal cord dMRI: From the upper cervical cord.
- Ex vivo specimen dMRI: From the temporal lobe of a patient.
These datasets were acquired using different scanner configurations and gradient schemes, highlighting ACID's versatility. The paper does not introduce new benchmarks but rather uses these datasets to showcase the application of ACID's pipelines.
Key Insights and Ideas
- Unified Approach for Diverse dMRI: ACID's strength lies in its ability to handle brain, spinal cord, and ex vivo dMRI data within a single, integrated framework, addressing the specific challenges of each.
- SPM Integration is Key: Leveraging SPM's established infrastructure provides ACID with powerful segmentation, spatial processing, and statistical analysis tools, making it a comprehensive solution from raw data to group-level statistics.
- BIDS Compliance for Standardization: Adherence to the BIDS standard facilitates data sharing, reproducibility, and interoperability with other neuroimaging tools.
- Advanced Artifact Correction is Crucial: The toolbox emphasizes state-of-the-art artifact correction (motion, eddy currents, susceptibility, Rician bias) as a prerequisite for accurate downstream modeling, especially for challenging data like spinal cord dMRI.
- Biophysical Models Need Validation: The inclusion of biophysical models highlights the growing need to link dMRI metrics to underlying tissue microstructure, with ex vivo data playing a vital role in validation.
Key Results
- ACID provides a comprehensive processing chain from raw DICOM to group-level statistics.
- The toolbox integrates advanced artifact correction methods (ECMOCO, msPOAS, RBC, HySCO) that are crucial for improving the accuracy of diffusion parameter estimation.
- ACID supports various diffusion signal models (DTI, DKI) and biophysical models (WMTI-Watson, NODDI-DTI).
- Example pipelines demonstrate ACID's application to in vivo brain, in vivo spinal cord, and ex vivo dMRI datasets, showcasing the output of various processing steps and model fits.
- The paper provides detailed discussions on the performance and considerations for each module, particularly for spinal cord dMRI and low-SNR scenarios.
Limitations and Future Work
- Parameter Optimization: The authors acknowledge that optimal processing parameters can vary significantly depending on the specific dMRI data, subject, and imaged region, encouraging users to perform systematic explorations on their own datasets.
- Comparison with Other Toolboxes: A comprehensive quantitative comparison between ACID and other dMRI toolboxes is warranted and suggested for future work.
- Biophysical Model Advancements: The field of biophysical modeling is rapidly evolving, and ACID aims to continuously integrate new methods.
- Rician Maximum Likelihood Estimator: The authors plan to add the Rician maximum likelihood estimator as an alternative to existing quasi-likelihood estimators.
- Clinical Use: ACID is intended for academic research and is not approved for clinical applications.
Summary updated